Automated Rockfall Feature Extraction using High-Resolution 3D Point Clouds

Rockfalls are critical landslide phenomena that significantly impact human activities. Many previous studies have struggled to quantify rockfall volumes due to challenges in volume estimation, particularly without modern remote sensing technologies. Traditional methods, such as those utilizing open-source software like CloudCompare to process 3D point cloud data from Terrestrial Laser Scanning (TLS), are often time-consuming and introduce considerable uncertainty in volume estimation. Moreover, the long-term volume and erosion rate changes of coastal cliffs are rarely addressed in detail.

This study focuses on evaluating rockfall hazards activity along active shoreline cliffs, specifically targeting a rock slope in the more than 20 km of the northern Yorkshire coast cliff, United Kingdom, where frequent rockfalls occur. Leveraging over 10 years of annual  high-resolution lidar data, we developed a rockfall database to assess erosion rates and volume changes over time. To streamline the analysis, we introduced a multi-phase processing framework unified into a single Python script, cobra.py. Preprocessing begins with raw data filtering, sampling, merging, and region-of-interest (ROI) extraction, guided by a shapefile prepared using geometric features, spatial relationships, and the verticality of the cliff face. The cobra.py script integrates consecutive analytical phases:

• Change Detection and Clustering: Eroded blocks and rockfall changes are identified using DBSCAN clustering and centroid proximity.
• Volume Estimation: 3D point cloud data are converted into voxel and mesh representations for accurate volume estimation of eroded blocks.
• Erosion Rate and Density Calculations: Poisson Surface Reconstruction is applied to calculate the cliff face area and consequently calculate the erosion rates.
• Cluster Shape Classification: Clusters are classified based on a tyranny plot of rock shape relationships, and outputs are visualized through plots and summary statistics.

Validation of the lidar-based inventory was performed using high-temporal-resolution TLS data collected at overlapping time periods and short sections of location. The estimated volumes and spatial correlations of rockfall blocks were assessed through descriptive statistics, empirical cumulative distribution functions (ECDF), and goodness-of-fit metrics. Differences in point cloud density and spatial matching errors were accounted for by increasing tolerance during validation. This developed integrated approach offers a robust framework for quantifying rockfall hazards and erosion processes, providing insights critical for coastal slope management and hazard mitigation.